2
$\begingroup$

Check the following, integer problem:

NMaximize[{3*0.5*(80 - P)*Subscript[Q, 1]*z1 + 
   2*0.5*(125 - P)*Subscript[Q, 2]*z2 + 
   1*0.5*(180 - 2*P)*Subscript[Q, 3]*z3,
  Subscript[Q, 1] == (80 - P) y1, 
  Subscript[Q, 2] == (100 - 0.8 P) y2, 
  Subscript[Q, 3] == (90 - 0.5 P) y3, 20 <= P <= 90, 1 >= y1 >= 0, 
  1 >= y2 >= 0, 1 >= y3 >= 0, 3 >= y1 + y2 + y3 >= 2, 1 >= z1 >= 0, 
  1 >= z2 >= 0, 1 >= z3 >= 0, z1 + z2 + z3 == 1, 
  3 <= y1 + y2 + y3 + z1 + z2 + z3 <= 4, y1 + y2 + y3 + z2 <= 3, 
  y1 + y2 + y3 + z3 <= 3, Subscript[Q, 1] >= 0, Subscript[Q, 2] >= 10,
   Subscript[Q, 3] >= 10, {y1, y2, y3, z1, z2, z3} \[Element] 
   Integers}, {P, Subscript[Q, 2], Subscript[Q, 3], Subscript[Q, 1], 
  y1, y2, y3, z1, z2, z3}]

Mathematica fails to find the optimal solution, which is

P->20, Q2->84, Q3->80, Y2->1, Y3->1, Z2->1

LINGO finds the global optimal solution automatically. I have tried with all methods provided in tutorial, but it did not help, unless I increased the lower bound to about Q2 > 60 and Q3 >60.

Any help?

$\endgroup$

3 Answers 3

2
$\begingroup$

I noticed that if I did not use Q1, Q2, or Q3 as intermediate values in the maximization then it worked:

NMaximize[{
  3*0.5*(80-P)*(80-P)*y1*z1 + 2*0.5*(125-P)*(100-0.8P)*y2*z2 + 1*0.5*(180-2P)*(90 - 0.5)*y3*z3,
  20 <= P <= 90,
  0 <= y1 <= 1,
  0 <= y2 <= 1,
  0 <= y3 <= 1,
  2 <= y1 + y2 + y3 <= 3,
  0 <= z1 <= 1,
  0 <= z2 <= 1,
  0 <= z3 <= 1,
  z1 + z2 + z3 == 1,
  3 <= y1 + y2 + y3 + z1 + z2 + z3 <= 4,
  y1 + y2 + y3 + z2 <= 3,
  y1 + y2 + y3 + z3 <= 3,
  (80 - P) y1 >= 0,
  (100 - 0.8 P) y2 >= 10,
  (90 - 0.5 P) y3 >= 10,
  {y1, y2, y3, z1, z2, z3} \[Element] Integers},
  {P, y1, y2, y3, z1, z2, z3}
]

This gave me the value:

{8820., {P->20, y1->0, y2->1, y3->1, z1->0, z2->1, z3->0}}

It appears that the constraints should only contain independent variables, and no dependent variables.

$\endgroup$
2
  • $\begingroup$ Michael, thanks very much for that. It helps a lot, but still the problem was not to find out the integer values, but the Q-values. I formulated it so that the integer (binary) will force to obtain the correct Q-values. I am surprised though that LINGO finds both the independent and dependent variables in a millisecond! $\endgroup$
    – AC Milan
    Commented Jan 16, 2019 at 9:14
  • $\begingroup$ @ACMilan max=NMaximize[...];{P, y1, y2, y3, z1, z2, z3} = {P, y1, y2, y3, z1, z2, z3} /. Last@max;Q1 = (80 - P) y1 Q2 = (100 - 0.8 P) y2 Q3 = (90 - 0.5 P) y3 gives exactly the LINGO's solution. $\endgroup$ Commented Jan 16, 2019 at 13:41
1
$\begingroup$

Avoid using subscripts!

NMaximize[{3*0.5*(80 - P)*Q1*z1 + 2*0.5*(125 -P)*Q2*z2 + 
1*0.5*(180 - 2*P)*Q3*z3, Q1 == (80 - P) y1, Q2 == 
(100 - 0.8 P) y2,
Q3 == (90 - 0.5 P) y3, 20 <= P <= 90, 1 >= y1 >= 0, 1 
>= y2 >= 0, 
1 >= y3 >= 0, 3 >= y1 + y2 + y3 >= 2, 1 >= z1 >= 0, 1 
>= z2 >= 0, 
1 >= z3 >= 0, z1 + z2 + z3 == 1, 
3  <= y1 + y2 + y3 + z1 + z2 + z3 <= 4, y1 + y2 + y3 
+ z2 <= 3, 
y1 + y2 + y3 + z3 <= 3, Q1 >= 0, Q2 >= 10, 
Q3 >= 10, {y1, y2, y3, z1, z2, z3} \[Element] 
Integers}, {P, Q2, Q3,
Q1, y1, y2, y3, z1, z2, z3}] 

(*{0.0741345, {P -> 79.7791, Q2 -> 10., Q3 -> 10., Q1-> 0.22375,y1 -> 1, y2 -> 0, y3 -> 0, z1 -> 1, z2 -> 0, z3 ->0}}*)
$\endgroup$
2
  • $\begingroup$ Ulrich,Subscripts make no difference; that is not the optimal solution! It is exactly the one NMaximize gives, which is wrong. $\endgroup$
    – AC Milan
    Commented Jan 16, 2019 at 9:27
  • $\begingroup$ @ACMilan Obviously we get a different solution without subscripts... $\endgroup$ Commented Jan 16, 2019 at 9:45
1
$\begingroup$

Since Binary domain is small we can use this approach.

pts = Select[
  Tuples[{0, 1}, 
   6], ((2 <= #[[1]] + #[[2]] + #[[3]] <= 
       3) && (#[[4]] + #[[5]] + #[[6]] == 1) && (3 <= Total@# <= 
       4) && (#[[1]] + #[[2]] + #[[3]] + #[[5]] <= 
       3) && (#[[1]] + #[[2]] + #[[3]] + #[[6]] <= 3)) &]

This gives All tuples with the constrains.

{y1, y2, y3, z1, z2, z3}={{0, 1, 1, 0, 0, 1}, {0, 1, 1, 0, 1, 0}, {0, 1, 1, 1, 0, 0}, {1, 0, 1, 0, 0, 1}, {1, 0, 1, 0, 1, 0}, {1, 0, 1, 1, 0, 0}, {1, 1, 0, 0, 0, 1}, {1, 1, 0, 0, 1, 0}, {1, 1, 0, 1, 0, 0}, {1, 1, 1, 1, 0, 0}}

Now we Maximize the obj func for each set of {y1, y2, y3, z1, z2, z3}

Table[ClearAll[y1, y2, y3, z1, z2, z3];
  {y1, y2, y3, z1, z2, z3} = pts[[i]];
  NMaximize[{3/2*(80 - P)*Q1*z1 + (125 - P)*Q2*z2 + 
     1/2*(180 - 2*P)*Q3*z3, Q1 == (80 - P) y1, Q2 == (100 - 4/5 P) y2,
     Q3 == (90 - 1/2 P) y3, 20 <= P <= 90, Q1 >= 0, Q2 >= 10, 
    Q3 >= 10}, {P, Q1, Q2, Q3}], {i, Length@pts}] // Quiet

{{5600., {P -> 20., Q1 -> 0, Q2 -> 84., Q3 -> 80.}}, {8820., {P -> 20., Q1 -> 0, Q2 -> 84., Q3 -> 80.}}, {0., {P -> 20.092, Q1 -> 0, Q2 -> 83.9264, Q3 -> 79.954}}, {-[Infinity], {P -> Indeterminate, Q1 -> Indeterminate, Q2 -> Indeterminate, Q3 -> Indeterminate}}, {-[Infinity], {P -> Indeterminate, Q1 -> Indeterminate, Q2 -> Indeterminate, Q3 -> Indeterminate}}, {-[Infinity], {P -> Indeterminate, Q1 -> Indeterminate, Q2 -> Indeterminate, Q3 -> Indeterminate}}, {-[Infinity], {P -> Indeterminate, Q1 -> Indeterminate, Q2 -> Indeterminate, Q3 -> Indeterminate}}, {-[Infinity], {P -> Indeterminate, Q1 -> Indeterminate, Q2 -> Indeterminate, Q3 -> Indeterminate}}, {-[Infinity], {P -> Indeterminate, Q1 -> Indeterminate, Q2 -> Indeterminate, Q3 -> Indeterminate}}, {5400., {P -> 20., Q1 -> 60., Q2 -> 84., Q3 -> 80.}}}

It is clear that Max value is 8820 with the values {P -> 20., Q1 -> 0, Q2 -> 84., Q3 -> 80.} which is second solution. So {y1, y2, y3, z1, z2, z3} ={0, 1, 1, 0, 1, 0}

$\endgroup$
2
  • $\begingroup$ Good approach (+1). I am not sure, though, is the OP question about (i) finding a solution with Mathematica or (ii) explaining or tweaking the behavior of NMaximize in order to find a solution. $\endgroup$ Commented Jan 16, 2019 at 15:22
  • $\begingroup$ Hi Anton. Exactly, my question was why NMaximize fails for such a small problem. I love Mathematica by the way! $\endgroup$
    – AC Milan
    Commented Jan 17, 2019 at 21:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.